1. Synaptic Energy Use and Supply
Julia J. Harris, Renaud Jolivet, David Attwell 
Neuron 
Volume 75, Issue 5, Pages (September 2012)
DOI: /j.neuron
Copyright © 2012 Elsevier Inc. Terms and Conditions

2. Figure 1 Mechanisms that Consume Energy at Synapses
ATP consumption by signaling mechanisms (Attwell and Laughlin, 2001). Presynaptically, ATP is used on four types of ATPase: the sodium pump, which extrudes Na+ ions generating the action potential and powers Ca2+ removal by Na+/Ca2+ exchange; calcium-ATPase in the plasma membrane (and endoplasmic reticulum, not shown), which lowers [Ca2+]i; vacuolar H+-ATPase, which energizes vesicle transmitter uptake; and motor proteins (kinesin, dynein, myosin) that move mitochondria and vesicles around the cell. In addition, vesicle retrieval by dynamin consumes GTP. Postsynaptically, ATP use is larger (shown by thicker arrows) and is mainly on the pumping out of ions mediating synaptic currents, with a smaller usage on returning Ca2+ to intracellular stores and on mitochondrial trafficking. In astrocytes ATP is used largely on extruding Na+, to maintain the resting potential and to remove the ions driving glutamate uptake, and on conversion of glutamate into glutamine. A small amount of energy (not included here) may also be consumed by signaling mediated by ATP.
Neuron  , DOI: ( /j.neuron )
Copyright © 2012 Elsevier Inc. Terms and Conditions

3. Figure 2 Energy Consumption of Different Signaling Mechanisms
(A) Percentage of ATP predicted to be used on the subcellular mechanisms of Figure 1, from the analysis of Attwell and Laughlin (2001), updated to account for mammalian action potentials using less energy than those in squid giant axon.
(B) The energy per vesicle released expended on different aspects of excitatory synaptic transmission (postsynaptic non-NMDA, NMDA, and metabotropic glutamate receptors; recycling of glutamate [glu], endo- and exocytosis of vesicles [endo/exo], and presynaptic Ca2+ entry) expressed as ATP molecules (left ordinate) or percentage (right ordinate). See Attwell and Laughlin (2001) for the derivation of these numbers.
(C) Distribution of mitochondria in different subcellular compartments observed by Wong-Riley (1989) in primate visual cortex, compared with the predicted distribution of energy expenditure from (A) with housekeeping energy added as 25% of the total (allocated in proportion to the volume of the different compartments, using neuronal dimensions from Attwell and Laughlin (2001) and an astrocyte volume of 1.49 × 10−14 m3 (Chvátal et al., 2007).
Neuron  , DOI: ( /j.neuron )
Copyright © 2012 Elsevier Inc. Terms and Conditions

4. Figure 3 Energy Limitations Imply that Synaptic Failures Are Desirable
Information flow from an input neuron (green) through a single synapse (to the orange neuron) or multiple synapses (to the brown neuron).
(A) Information transmitted along the input axon per time Δt (from Equation 1) is maximized at 1 bit/Δt when s, the probability of an action potential/Δt, is 0.5.
(B) Fraction of information arriving on the axon that is transmitted across a synapse releasing one vesicle with probability p, calculated as the mutual informationIm=Iinput(s)+ΣyP[y]·ΣxP[x|y]·log2(P[x|y])(where the input x is 1 when there is a spike and 0 otherwise, the output y is 1 when there is an EPSC and 0 otherwise, and the sum is over all x and y [Dayan and Abbott, 2001, Equation 4.12], which gives Equation 3), divided by the input information Iinput(s) in Equation 1.
(C) Ratio of information fraction transmitted (from B) to energy used on synaptic transmission, taken as s·p, the number of vesicles released per Δt, assuming that energy use is proportional to the number of vesicles released, with release of one vesicle consuming one unit of energy (if the small percentage (∼7%, Figure 2B) of ATP used on presynaptic Ca2+ pumping varies more weakly than p, the curves will decrease more as p approaches zero).
(D) Fraction of information arriving on an axon that is transmitted across N synapses each releasing one vesicle with probability p, in the absence of spontaneous release (Equation 4, black lines), and in the presence of mEPSCs occurring in the whole cell with a probability m = per Δt (Equation 5, red).
(E) Ratio of information fraction transmitted (from D) to postsynaptic energy used, taken as N·s·p (the number of vesicles released by action potentials per Δt), in the absence (black) and presence (red) of spontaneous release.
(F) Dependence on the factor r, by which spiking increases energy use, of the release probability that matches output information rate to the information arriving on a large number of independent input synapses (from Levy and Baxter, 2002). From Figure 2, per Δt = 2.5 ms, resting potentials and housekeeping use 3 × 106 ATP, while an action potential (and downstream synaptic events) uses 4.5 × 108 ATP and increases energy use r = 150-fold (arrow). For r = 150, Equation 2 predicts an optimal spike probability, to maximize axonal information transmission/energy used, of s∗ = (a firing rate of 9.7 Hz) and the graph predicts p ∼0.2.
Neuron  , DOI: ( /j.neuron )
Copyright © 2012 Elsevier Inc. Terms and Conditions

5. Figure 4 Energy limitations on number of postsynaptic receptors.
(A) Schematic dependence of the fraction of information transmitted by a relay synapse on the number of postsynaptic receptors per synapse (normalized to the number needed to generate an action potential). Below the threshold for action potential production no information is transmitted; above the threshold transmission is reliable.
(B) Ratio of fraction of information transmitted to energy used postsynaptically for the synapse of (A). Energy used is proportional to the number of receptors activated. Increasing the receptor number above the value needed to ensure transmission decreases the energy efficiency of the synapse. In practice the sharpness of the optimum in (B) will be degraded by noise.
(C) Dependence on the number of receptors in a postsynaptic spine, K, of the energy used on postsynaptic current (arbitrary units), the signal to noise ratio of the current (SN), and the signal-to-noise ratio to energy use (arbitrary units). The open probability of each activated receptor was assumed to be 0.5.
Neuron  , DOI: ( /j.neuron )
Copyright © 2012 Elsevier Inc. Terms and Conditions

6. Figure 5 Trafficking of Mitochondria to Synapses
Mitochondria move long distances along microtubules, driven by kinesin and dynein motors, and shorter distances along actin filaments, driven by myosin motors. Conversion of ATP to ADP by ATPases like the Na+ pump reduces the energy available for motor-driven transport, and ADP rebinding to the motor slows its movement. Calcium entering through presynaptic voltage-gated channels or postsynaptic NMDA receptors (or perhaps Ca2+-permeable AMPA/kainate receptors) binds to the adaptor protein Miro and stops kinesin motors moving the mitochondrion. Contrasting models for how this occurs are drawn pre- and postsynaptically. Once stopped, mitochondria may be tethered to microtubules by syntaphilin.
Neuron  , DOI: ( /j.neuron )
Copyright © 2012 Elsevier Inc. Terms and Conditions